Evaluation of The Digital Elevation Models Obtained From Göktürk-1 Satellite Data

Abstract

A Digital Elevation Model (DEM) is a model that uses the elevation information of the ground to define the land surface in three dimensions using mathematical equations. Because of its benefits, including speed of production, economy, lack of reliance on human labor, and simplicity of data access, satellite images are often chosen in the creation of DEMs. Furthermore, it is possible to attain great positional accuracy in both horizontal and vertical axes because of the metadata and Rational Polynomial Coefficients (RPC) that are included with the satellite images. Within the scope of the presented thesis, DEM was obtained using Göktürk-1 (GKT-1) stereo image pairs in three different regions in the world (United States of America (USA)/Texas-Amarillo, Italy/Monfalcone, Türkiye/Antalya). The elevation information of these models was compared separately with the open source ICESat-2 data and global digital elevation models ALOS DEM (AW3D30), Copernicus DEM, NASA DEM, and SRTM DEM (ACNS) data. During the comparison process, absolute mean, median, mean deviation, median absolute deviation, variance, and standard deviation statistical procedures were used. ICESat-2 data used in the thesis study was acquired taking into account the regions where Göktürk-1 stereo pairs coincide, and went through a series of filtering processes to be used as reference data. As a result of the filtering, 2629 ICESat-2 reference data in the USA/Texas-Amarillo region, 3489 in the Italy/Monfalcone region and 2215 in the Türkiye/Antalya region were used in the study. Sentinel-1 and Sentinel-2 based global land use and land cover (LULC) maps, WorldCover, of three different regions where Göktürk-1 stereo pairs are located were acquired from the European Space Agency (ESA) as open source. Changes in height differences between global DEM data and ICESat-2 data were examined according to the classes in the maps. During the height comparisons, absolute mean, median, absolute mean deviation, median absolute deviation, variance and standard deviation statistical analyzes were used. It has been observed that there are 8 different classes (Tree Cover, Shrubland, Grassland, Cropland, Built-up, Bare/Sparse Vegetation, Permanent Water Bodies, Herbaceous Wetlands) corresponding to Göktürk-1 DEM regions in the LULC data. Global digital elevation models were acquired to correspond to three different regions where Göktürk-1 stereo image pairs are located. In the thesis study, the comparison of each DEM with each other based on height (GKT1-ACNS, ICESat-2-ACNS, GKT1-ICESat-2) and all the results were presented. The average values found in the height comparison of ICESat-2 data with Global DEMs (ICESat-2-ACNS) in three regions in USA, Italy, and Türkiye are 2,553 / 1,415 / 1,854 and 2,877 meters, respectively. On the other hand, when the Göktürk-1 DEM without GCP (ICESat-2 - GKT-1 wo GCP) data, which corresponds to the ICESat-2 reference data, are compared, the average value emerges as 6,623 meters. As a result of the class-based comparison between ICESat-2 and GKT-1 DEM without GCP, it was seen that the results with the best accuracy were in the Italy/ Monfalcone region. The classes with the highest accuracy are shrubland/grassland/cropland classes with 2,102 / 2,135 / 2,118 meters respectively. DEM data with GCP was also produced from GKT-1 stereo images and compared with ICESat-2 reference data and Global DEM data. When ICESat-2 data are compared with GKT-1 DEM data with GCP (ICESat-2 – GKT-1 w GCP), the average value obtained for the three study areas is 1,807 meters. As a result of the class-based comparison between ICESat-2 and GKT-1 DEM with GCP, it was seen that the results with the best accuracy were in the USA/ Texas-Amarillo region. The classes with the highest accuracy are grassland/ cropland / bare-sparse vegetation classes with 0,693 / 0,665 / 0,635 meters respectively.